This project aims to develop a classification system designed to identify and categorize recipes based on specific dietary labels. These labels include whether a recipe is Vegetarian, Dairy Free, Gluten Free, Low Carb, Low Fat, and Low Sodium.
We worked on the Food Recipe Dataset available on Kaggle.
pip install -r requirements.txt./
├── 📂 project
│ ├── 📄 data_understanding.ipynb
│ ├── 📄 NER.ipynb
│ ├── 📄 classification_standard_ml.ipynb
│ ├── 📄 static_embeddings_classification.ipynb
│ ├── 📄 RNNs.ipynb
│ ├── 📄 finetuning_veg.ipynb
│ ├── 📄 finetuning_multi.ipynb
│ ├── 📄 requirements.txt
│ └── 📂 utils
│ ├── 📄 macro_tags.py
│ └── ...
│ └── 📂 old_notebooks
│ └── 📂 dataset
│ ├── 📄 dataset.csv
│ └── ...
│ └── 📂 models
│ ├── 📂 bert_model1
│ └── ...
├── 📄 README.md
└── 📄 .gitignore
data_understanding.ipynb: all the cells required to obtain the datasets needed for the rest of the project, and some data analysis.NER.ipynb: extraction of the ingredients and cleaning with spaCy and Stanza.classification_standard_ml.ipynb: experiments with RF, MLP, SVM with one-hot encoding.static_embeddings_classification.ipynb: experiments with RF, MLP, SVM, LogReg with Word2Vec.RNNs.ipynb: experiments with RNN architectures.finetuning_veg.ipynb: finetuning of BERT for the single-label task.finetuning_multi.ipynb: finetuning of BERT for the multi-label task.
The notebooks can be run in the listed order to replicate the experiments.
The report contains the description of the project, with the results of the most significant experiments, and bibliographic references.